xref: /dflybsd-src/sys/vfs/devfs/devfs_vnops.c (revision ad63697b1227b985f99dd96f67aaafb238b3d148)
1 /*
2  * (MPSAFE)
3  *
4  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Alex Hornung <ahornung@gmail.com>
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in
17  *    the documentation and/or other materials provided with the
18  *    distribution.
19  * 3. Neither the name of The DragonFly Project nor the names of its
20  *    contributors may be used to endorse or promote products derived
21  *    from this software without specific, prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/time.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/fcntl.h>
42 #include <sys/proc.h>
43 #include <sys/priv.h>
44 #include <sys/signalvar.h>
45 #include <sys/vnode.h>
46 #include <sys/uio.h>
47 #include <sys/mount.h>
48 #include <sys/file.h>
49 #include <sys/fcntl.h>
50 #include <sys/namei.h>
51 #include <sys/dirent.h>
52 #include <sys/malloc.h>
53 #include <sys/stat.h>
54 #include <sys/reg.h>
55 #include <vm/vm_pager.h>
56 #include <vm/vm_zone.h>
57 #include <vm/vm_object.h>
58 #include <sys/filio.h>
59 #include <sys/ttycom.h>
60 #include <sys/tty.h>
61 #include <sys/diskslice.h>
62 #include <sys/sysctl.h>
63 #include <sys/devfs.h>
64 #include <sys/pioctl.h>
65 #include <vfs/fifofs/fifo.h>
66 
67 #include <machine/limits.h>
68 
69 #include <sys/buf2.h>
70 #include <sys/sysref2.h>
71 #include <sys/mplock2.h>
72 #include <vm/vm_page2.h>
73 
74 MALLOC_DECLARE(M_DEVFS);
75 #define DEVFS_BADOP	(void *)devfs_vop_badop
76 
77 static int devfs_vop_badop(struct vop_generic_args *);
78 static int devfs_vop_access(struct vop_access_args *);
79 static int devfs_vop_inactive(struct vop_inactive_args *);
80 static int devfs_vop_reclaim(struct vop_reclaim_args *);
81 static int devfs_vop_readdir(struct vop_readdir_args *);
82 static int devfs_vop_getattr(struct vop_getattr_args *);
83 static int devfs_vop_setattr(struct vop_setattr_args *);
84 static int devfs_vop_readlink(struct vop_readlink_args *);
85 static int devfs_vop_print(struct vop_print_args *);
86 
87 static int devfs_vop_nresolve(struct vop_nresolve_args *);
88 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
89 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
90 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
91 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
92 static int devfs_vop_nremove(struct vop_nremove_args *);
93 
94 static int devfs_spec_open(struct vop_open_args *);
95 static int devfs_spec_close(struct vop_close_args *);
96 static int devfs_spec_fsync(struct vop_fsync_args *);
97 
98 static int devfs_spec_read(struct vop_read_args *);
99 static int devfs_spec_write(struct vop_write_args *);
100 static int devfs_spec_ioctl(struct vop_ioctl_args *);
101 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
102 static int devfs_spec_strategy(struct vop_strategy_args *);
103 static void devfs_spec_strategy_done(struct bio *);
104 static int devfs_spec_freeblks(struct vop_freeblks_args *);
105 static int devfs_spec_bmap(struct vop_bmap_args *);
106 static int devfs_spec_advlock(struct vop_advlock_args *);
107 static void devfs_spec_getpages_iodone(struct bio *);
108 static int devfs_spec_getpages(struct vop_getpages_args *);
109 
110 static int devfs_fo_close(struct file *);
111 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
112 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
113 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
114 static int devfs_fo_kqfilter(struct file *, struct knote *);
115 static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
116 				struct ucred *, struct sysmsg *);
117 static __inline int sequential_heuristic(struct uio *, struct file *);
118 
119 extern struct lock devfs_lock;
120 
121 /*
122  * devfs vnode operations for regular files.  All vnode ops are MPSAFE.
123  */
124 struct vop_ops devfs_vnode_norm_vops = {
125 	.vop_default =		vop_defaultop,
126 	.vop_access =		devfs_vop_access,
127 	.vop_advlock =		DEVFS_BADOP,
128 	.vop_bmap =		DEVFS_BADOP,
129 	.vop_close =		vop_stdclose,
130 	.vop_getattr =		devfs_vop_getattr,
131 	.vop_inactive =		devfs_vop_inactive,
132 	.vop_ncreate =		DEVFS_BADOP,
133 	.vop_nresolve =		devfs_vop_nresolve,
134 	.vop_nlookupdotdot =	devfs_vop_nlookupdotdot,
135 	.vop_nlink =		DEVFS_BADOP,
136 	.vop_nmkdir =		devfs_vop_nmkdir,
137 	.vop_nmknod =		DEVFS_BADOP,
138 	.vop_nremove =		devfs_vop_nremove,
139 	.vop_nrename =		DEVFS_BADOP,
140 	.vop_nrmdir =		devfs_vop_nrmdir,
141 	.vop_nsymlink =		devfs_vop_nsymlink,
142 	.vop_open =		vop_stdopen,
143 	.vop_pathconf =		vop_stdpathconf,
144 	.vop_print =		devfs_vop_print,
145 	.vop_read =		DEVFS_BADOP,
146 	.vop_readdir =		devfs_vop_readdir,
147 	.vop_readlink =		devfs_vop_readlink,
148 	.vop_reclaim =		devfs_vop_reclaim,
149 	.vop_setattr =		devfs_vop_setattr,
150 	.vop_write =		DEVFS_BADOP,
151 	.vop_ioctl =		DEVFS_BADOP
152 };
153 
154 /*
155  * devfs vnode operations for character devices.  All vnode ops are MPSAFE.
156  */
157 struct vop_ops devfs_vnode_dev_vops = {
158 	.vop_default =		vop_defaultop,
159 	.vop_access =		devfs_vop_access,
160 	.vop_advlock =		devfs_spec_advlock,
161 	.vop_bmap =		devfs_spec_bmap,
162 	.vop_close =		devfs_spec_close,
163 	.vop_freeblks =		devfs_spec_freeblks,
164 	.vop_fsync =		devfs_spec_fsync,
165 	.vop_getattr =		devfs_vop_getattr,
166 	.vop_getpages =		devfs_spec_getpages,
167 	.vop_inactive =		devfs_vop_inactive,
168 	.vop_open =		devfs_spec_open,
169 	.vop_pathconf =		vop_stdpathconf,
170 	.vop_print =		devfs_vop_print,
171 	.vop_kqfilter =		devfs_spec_kqfilter,
172 	.vop_read =		devfs_spec_read,
173 	.vop_readdir =		DEVFS_BADOP,
174 	.vop_readlink =		DEVFS_BADOP,
175 	.vop_reclaim =		devfs_vop_reclaim,
176 	.vop_setattr =		devfs_vop_setattr,
177 	.vop_strategy =		devfs_spec_strategy,
178 	.vop_write =		devfs_spec_write,
179 	.vop_ioctl =		devfs_spec_ioctl
180 };
181 
182 /*
183  * devfs file pointer operations.  All fileops are MPSAFE.
184  */
185 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
186 
187 struct fileops devfs_dev_fileops = {
188 	.fo_read	= devfs_fo_read,
189 	.fo_write	= devfs_fo_write,
190 	.fo_ioctl	= devfs_fo_ioctl,
191 	.fo_kqfilter	= devfs_fo_kqfilter,
192 	.fo_stat	= devfs_fo_stat,
193 	.fo_close	= devfs_fo_close,
194 	.fo_shutdown	= nofo_shutdown
195 };
196 
197 /*
198  * These two functions are possibly temporary hacks for devices (aka
199  * the pty code) which want to control the node attributes themselves.
200  *
201  * XXX we may ultimately desire to simply remove the uid/gid/mode
202  * from the node entirely.
203  *
204  * MPSAFE - sorta.  Theoretically the overwrite can compete since they
205  *	    are loading from the same fields.
206  */
207 static __inline void
208 node_sync_dev_get(struct devfs_node *node)
209 {
210 	cdev_t dev;
211 
212 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
213 		node->uid = dev->si_uid;
214 		node->gid = dev->si_gid;
215 		node->mode = dev->si_perms;
216 	}
217 }
218 
219 static __inline void
220 node_sync_dev_set(struct devfs_node *node)
221 {
222 	cdev_t dev;
223 
224 	if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
225 		dev->si_uid = node->uid;
226 		dev->si_gid = node->gid;
227 		dev->si_perms = node->mode;
228 	}
229 }
230 
231 /*
232  * generic entry point for unsupported operations
233  */
234 static int
235 devfs_vop_badop(struct vop_generic_args *ap)
236 {
237 	return (EIO);
238 }
239 
240 
241 static int
242 devfs_vop_access(struct vop_access_args *ap)
243 {
244 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
245 	int error;
246 
247 	if (!devfs_node_is_accessible(node))
248 		return ENOENT;
249 	node_sync_dev_get(node);
250 	error = vop_helper_access(ap, node->uid, node->gid,
251 				  node->mode, node->flags);
252 
253 	return error;
254 }
255 
256 
257 static int
258 devfs_vop_inactive(struct vop_inactive_args *ap)
259 {
260 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
261 
262 	if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
263 		vrecycle(ap->a_vp);
264 	return 0;
265 }
266 
267 
268 static int
269 devfs_vop_reclaim(struct vop_reclaim_args *ap)
270 {
271 	struct devfs_node *node;
272 	struct vnode *vp;
273 	int locked;
274 
275 	/*
276 	 * Check if it is locked already. if not, we acquire the devfs lock
277 	 */
278 	if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
279 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
280 		locked = 1;
281 	} else {
282 		locked = 0;
283 	}
284 
285 	/*
286 	 * Get rid of the devfs_node if it is no longer linked into the
287 	 * topology.
288 	 */
289 	vp = ap->a_vp;
290 	if ((node = DEVFS_NODE(vp)) != NULL) {
291 		node->v_node = NULL;
292 		if ((node->flags & DEVFS_NODE_LINKED) == 0)
293 			devfs_freep(node);
294 	}
295 
296 	if (locked)
297 		lockmgr(&devfs_lock, LK_RELEASE);
298 
299 	/*
300 	 * v_rdev needs to be properly released using v_release_rdev
301 	 * Make sure v_data is NULL as well.
302 	 */
303 	vp->v_data = NULL;
304 	v_release_rdev(vp);
305 	return 0;
306 }
307 
308 
309 static int
310 devfs_vop_readdir(struct vop_readdir_args *ap)
311 {
312 	struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
313 	struct devfs_node *node;
314 	int cookie_index;
315 	int ncookies;
316 	int error2;
317 	int error;
318 	int r;
319 	off_t *cookies;
320 	off_t saveoff;
321 
322 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
323 
324 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
325 		return (EINVAL);
326 	if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
327 		return (error);
328 
329 	if (!devfs_node_is_accessible(dnode)) {
330 		vn_unlock(ap->a_vp);
331 		return ENOENT;
332 	}
333 
334 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
335 
336 	saveoff = ap->a_uio->uio_offset;
337 
338 	if (ap->a_ncookies) {
339 		ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
340 		if (ncookies > 256)
341 			ncookies = 256;
342 		cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
343 		cookie_index = 0;
344 	} else {
345 		ncookies = -1;
346 		cookies = NULL;
347 		cookie_index = 0;
348 	}
349 
350 	nanotime(&dnode->atime);
351 
352 	if (saveoff == 0) {
353 		r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
354 				     DT_DIR, 1, ".");
355 		if (r)
356 			goto done;
357 		if (cookies)
358 			cookies[cookie_index] = saveoff;
359 		saveoff++;
360 		cookie_index++;
361 		if (cookie_index == ncookies)
362 			goto done;
363 	}
364 
365 	if (saveoff == 1) {
366 		if (dnode->parent) {
367 			r = vop_write_dirent(&error, ap->a_uio,
368 					     dnode->parent->d_dir.d_ino,
369 					     DT_DIR, 2, "..");
370 		} else {
371 			r = vop_write_dirent(&error, ap->a_uio,
372 					     dnode->d_dir.d_ino,
373 					     DT_DIR, 2, "..");
374 		}
375 		if (r)
376 			goto done;
377 		if (cookies)
378 			cookies[cookie_index] = saveoff;
379 		saveoff++;
380 		cookie_index++;
381 		if (cookie_index == ncookies)
382 			goto done;
383 	}
384 
385 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
386 		if ((node->flags & DEVFS_HIDDEN) ||
387 		    (node->flags & DEVFS_INVISIBLE)) {
388 			continue;
389 		}
390 
391 		/*
392 		 * If the node type is a valid devfs alias, then we make
393 		 * sure that the target isn't hidden. If it is, we don't
394 		 * show the link in the directory listing.
395 		 */
396 		if ((node->node_type == Plink) && (node->link_target != NULL) &&
397 			(node->link_target->flags & DEVFS_HIDDEN))
398 			continue;
399 
400 		if (node->cookie < saveoff)
401 			continue;
402 
403 		saveoff = node->cookie;
404 
405 		error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
406 					  node->d_dir.d_type,
407 					  node->d_dir.d_namlen,
408 					  node->d_dir.d_name);
409 
410 		if (error2)
411 			break;
412 
413 		saveoff++;
414 
415 		if (cookies)
416 			cookies[cookie_index] = node->cookie;
417 		++cookie_index;
418 		if (cookie_index == ncookies)
419 			break;
420 	}
421 
422 done:
423 	lockmgr(&devfs_lock, LK_RELEASE);
424 	vn_unlock(ap->a_vp);
425 
426 	ap->a_uio->uio_offset = saveoff;
427 	if (error && cookie_index == 0) {
428 		if (cookies) {
429 			kfree(cookies, M_TEMP);
430 			*ap->a_ncookies = 0;
431 			*ap->a_cookies = NULL;
432 		}
433 	} else {
434 		if (cookies) {
435 			*ap->a_ncookies = cookie_index;
436 			*ap->a_cookies = cookies;
437 		}
438 	}
439 	return (error);
440 }
441 
442 
443 static int
444 devfs_vop_nresolve(struct vop_nresolve_args *ap)
445 {
446 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
447 	struct devfs_node *node, *found = NULL;
448 	struct namecache *ncp;
449 	struct vnode *vp = NULL;
450 	int error = 0;
451 	int len;
452 	int depth;
453 
454 	ncp = ap->a_nch->ncp;
455 	len = ncp->nc_nlen;
456 
457 	if (!devfs_node_is_accessible(dnode))
458 		return ENOENT;
459 
460 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
461 
462 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) {
463 		error = ENOENT;
464 		cache_setvp(ap->a_nch, NULL);
465 		goto out;
466 	}
467 
468 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
469 		if (len == node->d_dir.d_namlen) {
470 			if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
471 				found = node;
472 				break;
473 			}
474 		}
475 	}
476 
477 	if (found) {
478 		depth = 0;
479 		while ((found->node_type == Plink) && (found->link_target)) {
480 			if (depth >= 8) {
481 				devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
482 				break;
483 			}
484 
485 			found = found->link_target;
486 			++depth;
487 		}
488 
489 		if (!(found->flags & DEVFS_HIDDEN))
490 			devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
491 	}
492 
493 	if (vp == NULL) {
494 		error = ENOENT;
495 		cache_setvp(ap->a_nch, NULL);
496 		goto out;
497 
498 	}
499 	KKASSERT(vp);
500 	vn_unlock(vp);
501 	cache_setvp(ap->a_nch, vp);
502 	vrele(vp);
503 out:
504 	lockmgr(&devfs_lock, LK_RELEASE);
505 
506 	return error;
507 }
508 
509 
510 static int
511 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
512 {
513 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
514 
515 	*ap->a_vpp = NULL;
516 	if (!devfs_node_is_accessible(dnode))
517 		return ENOENT;
518 
519 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
520 	if (dnode->parent != NULL) {
521 		devfs_allocv(ap->a_vpp, dnode->parent);
522 		vn_unlock(*ap->a_vpp);
523 	}
524 	lockmgr(&devfs_lock, LK_RELEASE);
525 
526 	return ((*ap->a_vpp == NULL) ? ENOENT : 0);
527 }
528 
529 
530 static int
531 devfs_vop_getattr(struct vop_getattr_args *ap)
532 {
533 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
534 	struct vattr *vap = ap->a_vap;
535 	struct partinfo pinfo;
536 	int error = 0;
537 
538 #if 0
539 	if (!devfs_node_is_accessible(node))
540 		return ENOENT;
541 #endif
542 	node_sync_dev_get(node);
543 
544 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
545 
546 	/* start by zeroing out the attributes */
547 	VATTR_NULL(vap);
548 
549 	/* next do all the common fields */
550 	vap->va_type = ap->a_vp->v_type;
551 	vap->va_mode = node->mode;
552 	vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
553 	vap->va_flags = 0;
554 	vap->va_blocksize = DEV_BSIZE;
555 	vap->va_bytes = vap->va_size = 0;
556 
557 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
558 
559 	vap->va_atime = node->atime;
560 	vap->va_mtime = node->mtime;
561 	vap->va_ctime = node->ctime;
562 
563 	vap->va_nlink = 1; /* number of references to file */
564 
565 	vap->va_uid = node->uid;
566 	vap->va_gid = node->gid;
567 
568 	vap->va_rmajor = 0;
569 	vap->va_rminor = 0;
570 
571 	if ((node->node_type == Pdev) && node->d_dev)  {
572 		reference_dev(node->d_dev);
573 		vap->va_rminor = node->d_dev->si_uminor;
574 		release_dev(node->d_dev);
575 	}
576 
577 	/* For a softlink the va_size is the length of the softlink */
578 	if (node->symlink_name != 0) {
579 		vap->va_bytes = vap->va_size = node->symlink_namelen;
580 	}
581 
582 	/*
583 	 * For a disk-type device, va_size is the size of the underlying
584 	 * device, so that lseek() works properly.
585 	 */
586 	if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
587 		bzero(&pinfo, sizeof(pinfo));
588 		error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
589 				   0, proc0.p_ucred, NULL);
590 		if ((error == 0) && (pinfo.media_blksize != 0)) {
591 			vap->va_size = pinfo.media_size;
592 		} else {
593 			vap->va_size = 0;
594 			error = 0;
595 		}
596 	}
597 
598 	lockmgr(&devfs_lock, LK_RELEASE);
599 
600 	return (error);
601 }
602 
603 
604 static int
605 devfs_vop_setattr(struct vop_setattr_args *ap)
606 {
607 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
608 	struct vattr *vap;
609 	uid_t cur_uid;
610 	gid_t cur_gid;
611 	mode_t cur_mode;
612 	int error = 0;
613 
614 	if (!devfs_node_is_accessible(node))
615 		return ENOENT;
616 	node_sync_dev_get(node);
617 
618 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
619 
620 	vap = ap->a_vap;
621 
622 	if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
623 		cur_uid = node->uid;
624 		cur_gid = node->gid;
625 		cur_mode = node->mode;
626 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
627 		    ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
628 		if (error)
629 			goto out;
630 
631 		if (node->uid != cur_uid || node->gid != cur_gid) {
632 			node->uid = cur_uid;
633 			node->gid = cur_gid;
634 			node->mode = cur_mode;
635 		}
636 	}
637 
638 	if (vap->va_mode != (mode_t)VNOVAL) {
639 		cur_mode = node->mode;
640 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
641 		    node->uid, node->gid, &cur_mode);
642 		if (error == 0 && node->mode != cur_mode) {
643 			node->mode = cur_mode;
644 		}
645 	}
646 
647 out:
648 	node_sync_dev_set(node);
649 	nanotime(&node->ctime);
650 	lockmgr(&devfs_lock, LK_RELEASE);
651 
652 	return error;
653 }
654 
655 
656 static int
657 devfs_vop_readlink(struct vop_readlink_args *ap)
658 {
659 	struct devfs_node *node = DEVFS_NODE(ap->a_vp);
660 	int ret;
661 
662 	if (!devfs_node_is_accessible(node))
663 		return ENOENT;
664 
665 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
666 	ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
667 	lockmgr(&devfs_lock, LK_RELEASE);
668 
669 	return ret;
670 }
671 
672 
673 static int
674 devfs_vop_print(struct vop_print_args *ap)
675 {
676 	return (0);
677 }
678 
679 static int
680 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
681 {
682 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
683 	struct devfs_node *node;
684 
685 	if (!devfs_node_is_accessible(dnode))
686 		return ENOENT;
687 
688 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
689 		goto out;
690 
691 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
692 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Pdir,
693 		      ap->a_nch->ncp->nc_name, dnode, NULL);
694 
695 	if (*ap->a_vpp) {
696 		node = DEVFS_NODE(*ap->a_vpp);
697 		node->flags |= DEVFS_USER_CREATED;
698 		cache_setunresolved(ap->a_nch);
699 		cache_setvp(ap->a_nch, *ap->a_vpp);
700 	}
701 	lockmgr(&devfs_lock, LK_RELEASE);
702 out:
703 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
704 }
705 
706 static int
707 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
708 {
709 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
710 	struct devfs_node *node;
711 	size_t targetlen;
712 
713 	if (!devfs_node_is_accessible(dnode))
714 		return ENOENT;
715 
716 	ap->a_vap->va_type = VLNK;
717 
718 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
719 		goto out;
720 
721 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
722 	devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink,
723 		      ap->a_nch->ncp->nc_name, dnode, NULL);
724 
725 	targetlen = strlen(ap->a_target);
726 	if (*ap->a_vpp) {
727 		node = DEVFS_NODE(*ap->a_vpp);
728 		node->flags |= DEVFS_USER_CREATED;
729 		node->symlink_namelen = targetlen;
730 		node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
731 		memcpy(node->symlink_name, ap->a_target, targetlen);
732 		node->symlink_name[targetlen] = '\0';
733 		cache_setunresolved(ap->a_nch);
734 		cache_setvp(ap->a_nch, *ap->a_vpp);
735 	}
736 	lockmgr(&devfs_lock, LK_RELEASE);
737 out:
738 	return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
739 }
740 
741 static int
742 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
743 {
744 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
745 	struct devfs_node *node;
746 	struct namecache *ncp;
747 	int error = ENOENT;
748 
749 	ncp = ap->a_nch->ncp;
750 
751 	if (!devfs_node_is_accessible(dnode))
752 		return ENOENT;
753 
754 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
755 
756 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
757 		goto out;
758 
759 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
760 		if (ncp->nc_nlen != node->d_dir.d_namlen)
761 			continue;
762 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
763 			continue;
764 
765 		/*
766 		 * only allow removal of user created dirs
767 		 */
768 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
769 			error = EPERM;
770 			goto out;
771 		} else if (node->node_type != Pdir) {
772 			error = ENOTDIR;
773 			goto out;
774 		} else if (node->nchildren > 2) {
775 			error = ENOTEMPTY;
776 			goto out;
777 		} else {
778 			if (node->v_node)
779 				cache_inval_vp(node->v_node, CINV_DESTROY);
780 			devfs_unlinkp(node);
781 			error = 0;
782 			break;
783 		}
784 	}
785 
786 	cache_setunresolved(ap->a_nch);
787 	cache_setvp(ap->a_nch, NULL);
788 
789 out:
790 	lockmgr(&devfs_lock, LK_RELEASE);
791 	return error;
792 }
793 
794 static int
795 devfs_vop_nremove(struct vop_nremove_args *ap)
796 {
797 	struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
798 	struct devfs_node *node;
799 	struct namecache *ncp;
800 	int error = ENOENT;
801 
802 	ncp = ap->a_nch->ncp;
803 
804 	if (!devfs_node_is_accessible(dnode))
805 		return ENOENT;
806 
807 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
808 
809 	if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
810 		goto out;
811 
812 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
813 		if (ncp->nc_nlen != node->d_dir.d_namlen)
814 			continue;
815 		if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
816 			continue;
817 
818 		/*
819 		 * only allow removal of user created stuff (e.g. symlinks)
820 		 */
821 		if ((node->flags & DEVFS_USER_CREATED) == 0) {
822 			error = EPERM;
823 			goto out;
824 		} else if (node->node_type == Pdir) {
825 			error = EISDIR;
826 			goto out;
827 		} else {
828 			if (node->v_node)
829 				cache_inval_vp(node->v_node, CINV_DESTROY);
830 			devfs_unlinkp(node);
831 			error = 0;
832 			break;
833 		}
834 	}
835 
836 	cache_setunresolved(ap->a_nch);
837 	cache_setvp(ap->a_nch, NULL);
838 
839 out:
840 	lockmgr(&devfs_lock, LK_RELEASE);
841 	return error;
842 }
843 
844 
845 static int
846 devfs_spec_open(struct vop_open_args *ap)
847 {
848 	struct vnode *vp = ap->a_vp;
849 	struct vnode *orig_vp = NULL;
850 	struct devfs_node *node = DEVFS_NODE(vp);
851 	struct devfs_node *newnode;
852 	cdev_t dev, ndev = NULL;
853 	int error = 0;
854 
855 	if (node) {
856 		if (node->d_dev == NULL)
857 			return ENXIO;
858 		if (!devfs_node_is_accessible(node))
859 			return ENOENT;
860 	}
861 
862 	if ((dev = vp->v_rdev) == NULL)
863 		return ENXIO;
864 
865 	if (node && ap->a_fp) {
866 		devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
867 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
868 
869 		ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
870 						ap->a_mode, ap->a_cred);
871 		if (ndev != NULL) {
872 			newnode = devfs_create_device_node(
873 					DEVFS_MNTDATA(vp->v_mount)->root_node,
874 					ndev, NULL, NULL);
875 			/* XXX: possibly destroy device if this happens */
876 
877 			if (newnode != NULL) {
878 				dev = ndev;
879 				devfs_link_dev(dev);
880 
881 				devfs_debug(DEVFS_DEBUG_DEBUG,
882 						"parent here is: %s, node is: |%s|\n",
883 						((node->parent->node_type == Proot) ?
884 						"ROOT!" : node->parent->d_dir.d_name),
885 						newnode->d_dir.d_name);
886 				devfs_debug(DEVFS_DEBUG_DEBUG,
887 						"test: %s\n",
888 						((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
889 
890 				/*
891 				 * orig_vp is set to the original vp if we cloned.
892 				 */
893 				/* node->flags |= DEVFS_CLONED; */
894 				devfs_allocv(&vp, newnode);
895 				orig_vp = ap->a_vp;
896 				ap->a_vp = vp;
897 			}
898 		}
899 		lockmgr(&devfs_lock, LK_RELEASE);
900 	}
901 
902 	devfs_debug(DEVFS_DEBUG_DEBUG,
903 		    "devfs_spec_open() called on %s! \n",
904 		    dev->si_name);
905 
906 	/*
907 	 * Make this field valid before any I/O in ->d_open
908 	 */
909 	if (!dev->si_iosize_max)
910 		/* XXX: old DFLTPHYS == 64KB dependency */
911 		dev->si_iosize_max = min(MAXPHYS,64*1024);
912 
913 	if (dev_dflags(dev) & D_TTY)
914 		vsetflags(vp, VISTTY);
915 
916 	/*
917 	 * Open underlying device
918 	 */
919 	vn_unlock(vp);
920 	error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
921 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
922 
923 	/*
924 	 * Clean up any cloned vp if we error out.
925 	 */
926 	if (error) {
927 		if (orig_vp) {
928 			vput(vp);
929 			ap->a_vp = orig_vp;
930 			/* orig_vp = NULL; */
931 		}
932 		return error;
933 	}
934 
935 	/*
936 	 * This checks if the disk device is going to be opened for writing.
937 	 * It will be only allowed in the cases where securelevel permits it
938 	 * and it's not mounted R/W.
939 	 */
940 	if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
941 	    (ap->a_cred != FSCRED)) {
942 
943 		/* Very secure mode. No open for writing allowed */
944 		if (securelevel >= 2)
945 			return EPERM;
946 
947 		/*
948 		 * If it is mounted R/W, do not allow to open for writing.
949 		 * In the case it's mounted read-only but securelevel
950 		 * is >= 1, then do not allow opening for writing either.
951 		 */
952 		if (vfs_mountedon(vp)) {
953 			if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
954 				return EBUSY;
955 			else if (securelevel >= 1)
956 				return EPERM;
957 		}
958 	}
959 
960 	if (dev_dflags(dev) & D_TTY) {
961 		if (dev->si_tty) {
962 			struct tty *tp;
963 			tp = dev->si_tty;
964 			if (!tp->t_stop) {
965 				devfs_debug(DEVFS_DEBUG_DEBUG,
966 					    "devfs: no t_stop\n");
967 				tp->t_stop = nottystop;
968 			}
969 		}
970 	}
971 
972 
973 	if (vn_isdisk(vp, NULL)) {
974 		if (!dev->si_bsize_phys)
975 			dev->si_bsize_phys = DEV_BSIZE;
976 		vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
977 	}
978 
979 	vop_stdopen(ap);
980 #if 0
981 	if (node)
982 		nanotime(&node->atime);
983 #endif
984 
985 	/*
986 	 * If we replaced the vp the vop_stdopen() call will have loaded
987 	 * it into fp->f_data and vref()d the vp, giving us two refs.  So
988 	 * instead of just unlocking it here we have to vput() it.
989 	 */
990 	if (orig_vp)
991 		vput(vp);
992 
993 	/* Ugly pty magic, to make pty devices appear once they are opened */
994 	if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
995 		node->flags &= ~DEVFS_INVISIBLE;
996 
997 	if (ap->a_fp) {
998 		KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
999 		KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
1000 		ap->a_fp->f_ops = &devfs_dev_fileops;
1001 		KKASSERT(ap->a_fp->f_data == (void *)vp);
1002 	}
1003 
1004 	return 0;
1005 }
1006 
1007 
1008 static int
1009 devfs_spec_close(struct vop_close_args *ap)
1010 {
1011 	struct devfs_node *node;
1012 	struct proc *p = curproc;
1013 	struct vnode *vp = ap->a_vp;
1014 	cdev_t dev = vp->v_rdev;
1015 	int error = 0;
1016 	int needrelock;
1017 
1018 	if (dev)
1019 		devfs_debug(DEVFS_DEBUG_DEBUG,
1020 			    "devfs_spec_close() called on %s! \n",
1021 			    dev->si_name);
1022 	else
1023 		devfs_debug(DEVFS_DEBUG_DEBUG,
1024 			    "devfs_spec_close() called, null vode!\n");
1025 
1026 	/*
1027 	 * A couple of hacks for devices and tty devices.  The
1028 	 * vnode ref count cannot be used to figure out the
1029 	 * last close, but we can use v_opencount now that
1030 	 * revoke works properly.
1031 	 *
1032 	 * Detect the last close on a controlling terminal and clear
1033 	 * the session (half-close).
1034 	 */
1035 	if (dev)
1036 		reference_dev(dev);
1037 
1038 	if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1039 		p->p_session->s_ttyvp = NULL;
1040 		vrele(vp);
1041 	}
1042 
1043 	/*
1044 	 * Vnodes can be opened and closed multiple times.  Do not really
1045 	 * close the device unless (1) it is being closed forcibly,
1046 	 * (2) the device wants to track closes, or (3) this is the last
1047 	 * vnode doing its last close on the device.
1048 	 *
1049 	 * XXX the VXLOCK (force close) case can leave vnodes referencing
1050 	 * a closed device.  This might not occur now that our revoke is
1051 	 * fixed.
1052 	 */
1053 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1054 	if (dev && ((vp->v_flag & VRECLAIMED) ||
1055 	    (dev_dflags(dev) & D_TRACKCLOSE) ||
1056 	    (vp->v_opencount == 1))) {
1057 		/*
1058 		 * Ugly pty magic, to make pty devices disappear again once
1059 		 * they are closed.
1060 		 */
1061 		node = DEVFS_NODE(ap->a_vp);
1062 		if (node && (node->flags & DEVFS_PTY))
1063 			node->flags |= DEVFS_INVISIBLE;
1064 
1065 		/*
1066 		 * Unlock around dev_dclose(), unless the vnode is
1067 		 * undergoing a vgone/reclaim (during umount).
1068 		 */
1069 		needrelock = 0;
1070 		if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
1071 			needrelock = 1;
1072 			vn_unlock(vp);
1073 		}
1074 
1075 		/*
1076 		 * WARNING!  If the device destroys itself the devfs node
1077 		 *	     can disappear here.
1078 		 *
1079 		 * WARNING!  vn_lock() will fail if the vp is in a VRECLAIM,
1080 		 *	     which can occur during umount.
1081 		 */
1082 		error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
1083 		/* node is now stale */
1084 
1085 		if (needrelock) {
1086 			if (vn_lock(vp, LK_EXCLUSIVE | LK_RETRY) != 0) {
1087 				panic("devfs_spec_close: vnode %p "
1088 				      "unexpectedly could not be relocked",
1089 				      vp);
1090 			}
1091 		}
1092 	} else {
1093 		error = 0;
1094 	}
1095 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1096 
1097 	/*
1098 	 * Track the actual opens and closes on the vnode.  The last close
1099 	 * disassociates the rdev.  If the rdev is already disassociated or
1100 	 * the opencount is already 0, the vnode might have been revoked
1101 	 * and no further opencount tracking occurs.
1102 	 */
1103 	if (dev)
1104 		release_dev(dev);
1105 	if (vp->v_opencount > 0)
1106 		vop_stdclose(ap);
1107 	return(error);
1108 
1109 }
1110 
1111 
1112 static int
1113 devfs_fo_close(struct file *fp)
1114 {
1115 	struct vnode *vp = (struct vnode *)fp->f_data;
1116 	int error;
1117 
1118 	fp->f_ops = &badfileops;
1119 	error = vn_close(vp, fp->f_flag);
1120 
1121 	return (error);
1122 }
1123 
1124 
1125 /*
1126  * Device-optimized file table vnode read routine.
1127  *
1128  * This bypasses the VOP table and talks directly to the device.  Most
1129  * filesystems just route to specfs and can make this optimization.
1130  *
1131  * MPALMOSTSAFE - acquires mplock
1132  */
1133 static int
1134 devfs_fo_read(struct file *fp, struct uio *uio,
1135 		 struct ucred *cred, int flags)
1136 {
1137 	struct devfs_node *node;
1138 	struct vnode *vp;
1139 	int ioflag;
1140 	int error;
1141 	cdev_t dev;
1142 
1143 	KASSERT(uio->uio_td == curthread,
1144 		("uio_td %p is not td %p", uio->uio_td, curthread));
1145 
1146 	if (uio->uio_resid == 0)
1147 		return 0;
1148 
1149 	vp = (struct vnode *)fp->f_data;
1150 	if (vp == NULL || vp->v_type == VBAD)
1151 		return EBADF;
1152 
1153 	node = DEVFS_NODE(vp);
1154 
1155 	if ((dev = vp->v_rdev) == NULL)
1156 		return EBADF;
1157 
1158 	reference_dev(dev);
1159 
1160 	if ((flags & O_FOFFSET) == 0)
1161 		uio->uio_offset = fp->f_offset;
1162 
1163 	ioflag = 0;
1164 	if (flags & O_FBLOCKING) {
1165 		/* ioflag &= ~IO_NDELAY; */
1166 	} else if (flags & O_FNONBLOCKING) {
1167 		ioflag |= IO_NDELAY;
1168 	} else if (fp->f_flag & FNONBLOCK) {
1169 		ioflag |= IO_NDELAY;
1170 	}
1171 	if (flags & O_FBUFFERED) {
1172 		/* ioflag &= ~IO_DIRECT; */
1173 	} else if (flags & O_FUNBUFFERED) {
1174 		ioflag |= IO_DIRECT;
1175 	} else if (fp->f_flag & O_DIRECT) {
1176 		ioflag |= IO_DIRECT;
1177 	}
1178 	ioflag |= sequential_heuristic(uio, fp);
1179 
1180 	error = dev_dread(dev, uio, ioflag);
1181 
1182 	release_dev(dev);
1183 	if (node)
1184 		nanotime(&node->atime);
1185 	if ((flags & O_FOFFSET) == 0)
1186 		fp->f_offset = uio->uio_offset;
1187 	fp->f_nextoff = uio->uio_offset;
1188 
1189 	return (error);
1190 }
1191 
1192 
1193 static int
1194 devfs_fo_write(struct file *fp, struct uio *uio,
1195 		  struct ucred *cred, int flags)
1196 {
1197 	struct devfs_node *node;
1198 	struct vnode *vp;
1199 	int ioflag;
1200 	int error;
1201 	cdev_t dev;
1202 
1203 	KASSERT(uio->uio_td == curthread,
1204 		("uio_td %p is not p %p", uio->uio_td, curthread));
1205 
1206 	vp = (struct vnode *)fp->f_data;
1207 	if (vp == NULL || vp->v_type == VBAD)
1208 		return EBADF;
1209 
1210 	node = DEVFS_NODE(vp);
1211 
1212 	if (vp->v_type == VREG)
1213 		bwillwrite(uio->uio_resid);
1214 
1215 	vp = (struct vnode *)fp->f_data;
1216 
1217 	if ((dev = vp->v_rdev) == NULL)
1218 		return EBADF;
1219 
1220 	reference_dev(dev);
1221 
1222 	if ((flags & O_FOFFSET) == 0)
1223 		uio->uio_offset = fp->f_offset;
1224 
1225 	ioflag = IO_UNIT;
1226 	if (vp->v_type == VREG &&
1227 	   ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1228 		ioflag |= IO_APPEND;
1229 	}
1230 
1231 	if (flags & O_FBLOCKING) {
1232 		/* ioflag &= ~IO_NDELAY; */
1233 	} else if (flags & O_FNONBLOCKING) {
1234 		ioflag |= IO_NDELAY;
1235 	} else if (fp->f_flag & FNONBLOCK) {
1236 		ioflag |= IO_NDELAY;
1237 	}
1238 	if (flags & O_FBUFFERED) {
1239 		/* ioflag &= ~IO_DIRECT; */
1240 	} else if (flags & O_FUNBUFFERED) {
1241 		ioflag |= IO_DIRECT;
1242 	} else if (fp->f_flag & O_DIRECT) {
1243 		ioflag |= IO_DIRECT;
1244 	}
1245 	if (flags & O_FASYNCWRITE) {
1246 		/* ioflag &= ~IO_SYNC; */
1247 	} else if (flags & O_FSYNCWRITE) {
1248 		ioflag |= IO_SYNC;
1249 	} else if (fp->f_flag & O_FSYNC) {
1250 		ioflag |= IO_SYNC;
1251 	}
1252 
1253 	if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1254 		ioflag |= IO_SYNC;
1255 	ioflag |= sequential_heuristic(uio, fp);
1256 
1257 	error = dev_dwrite(dev, uio, ioflag);
1258 
1259 	release_dev(dev);
1260 	if (node) {
1261 		nanotime(&node->atime);
1262 		nanotime(&node->mtime);
1263 	}
1264 
1265 	if ((flags & O_FOFFSET) == 0)
1266 		fp->f_offset = uio->uio_offset;
1267 	fp->f_nextoff = uio->uio_offset;
1268 
1269 	return (error);
1270 }
1271 
1272 
1273 static int
1274 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1275 {
1276 	struct vnode *vp;
1277 	struct vattr vattr;
1278 	struct vattr *vap;
1279 	u_short mode;
1280 	cdev_t dev;
1281 	int error;
1282 
1283 	vp = (struct vnode *)fp->f_data;
1284 	if (vp == NULL || vp->v_type == VBAD)
1285 		return EBADF;
1286 
1287 	error = vn_stat(vp, sb, cred);
1288 	if (error)
1289 		return (error);
1290 
1291 	vap = &vattr;
1292 	error = VOP_GETATTR(vp, vap);
1293 	if (error)
1294 		return (error);
1295 
1296 	/*
1297 	 * Zero the spare stat fields
1298 	 */
1299 	sb->st_lspare = 0;
1300 	sb->st_qspare1 = 0;
1301 	sb->st_qspare2 = 0;
1302 
1303 	/*
1304 	 * Copy from vattr table ... or not in case it's a cloned device
1305 	 */
1306 	if (vap->va_fsid != VNOVAL)
1307 		sb->st_dev = vap->va_fsid;
1308 	else
1309 		sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1310 
1311 	sb->st_ino = vap->va_fileid;
1312 
1313 	mode = vap->va_mode;
1314 	mode |= S_IFCHR;
1315 	sb->st_mode = mode;
1316 
1317 	if (vap->va_nlink > (nlink_t)-1)
1318 		sb->st_nlink = (nlink_t)-1;
1319 	else
1320 		sb->st_nlink = vap->va_nlink;
1321 
1322 	sb->st_uid = vap->va_uid;
1323 	sb->st_gid = vap->va_gid;
1324 	sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1325 	sb->st_size = vap->va_bytes;
1326 	sb->st_atimespec = vap->va_atime;
1327 	sb->st_mtimespec = vap->va_mtime;
1328 	sb->st_ctimespec = vap->va_ctime;
1329 
1330 	/*
1331 	 * A VCHR and VBLK device may track the last access and last modified
1332 	 * time independantly of the filesystem.  This is particularly true
1333 	 * because device read and write calls may bypass the filesystem.
1334 	 */
1335 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
1336 		dev = vp->v_rdev;
1337 		if (dev != NULL) {
1338 			if (dev->si_lastread) {
1339 				sb->st_atimespec.tv_sec = dev->si_lastread;
1340 				sb->st_atimespec.tv_nsec = 0;
1341 			}
1342 			if (dev->si_lastwrite) {
1343 				sb->st_atimespec.tv_sec = dev->si_lastwrite;
1344 				sb->st_atimespec.tv_nsec = 0;
1345 			}
1346 		}
1347 	}
1348 
1349         /*
1350 	 * According to www.opengroup.org, the meaning of st_blksize is
1351 	 *   "a filesystem-specific preferred I/O block size for this
1352 	 *    object.  In some filesystem types, this may vary from file
1353 	 *    to file"
1354 	 * Default to PAGE_SIZE after much discussion.
1355 	 */
1356 
1357 	sb->st_blksize = PAGE_SIZE;
1358 
1359 	sb->st_flags = vap->va_flags;
1360 
1361 	error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1362 	if (error)
1363 		sb->st_gen = 0;
1364 	else
1365 		sb->st_gen = (u_int32_t)vap->va_gen;
1366 
1367 	sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1368 
1369 	return (0);
1370 }
1371 
1372 
1373 static int
1374 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1375 {
1376 	struct vnode *vp;
1377 	int error;
1378 	cdev_t dev;
1379 
1380 	vp = (struct vnode *)fp->f_data;
1381 	if (vp == NULL || vp->v_type == VBAD) {
1382 		error = EBADF;
1383 		goto done;
1384 	}
1385 	if ((dev = vp->v_rdev) == NULL) {
1386 		error = EBADF;
1387 		goto done;
1388 	}
1389 	reference_dev(dev);
1390 
1391 	error = dev_dkqfilter(dev, kn);
1392 
1393 	release_dev(dev);
1394 
1395 done:
1396 	return (error);
1397 }
1398 
1399 /*
1400  * MPALMOSTSAFE - acquires mplock
1401  */
1402 static int
1403 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1404 		  struct ucred *ucred, struct sysmsg *msg)
1405 {
1406 	struct devfs_node *node;
1407 	struct vnode *vp;
1408 	struct vnode *ovp;
1409 	cdev_t	dev;
1410 	int error;
1411 	struct fiodname_args *name_args;
1412 	size_t namlen;
1413 	const char *name;
1414 
1415 	vp = ((struct vnode *)fp->f_data);
1416 
1417 	if ((dev = vp->v_rdev) == NULL)
1418 		return EBADF;		/* device was revoked */
1419 
1420 	reference_dev(dev);
1421 
1422 	node = DEVFS_NODE(vp);
1423 
1424 	devfs_debug(DEVFS_DEBUG_DEBUG,
1425 		    "devfs_fo_ioctl() called! for dev %s\n",
1426 		    dev->si_name);
1427 
1428 	if (com == FIODTYPE) {
1429 		*(int *)data = dev_dflags(dev) & D_TYPEMASK;
1430 		error = 0;
1431 		goto out;
1432 	} else if (com == FIODNAME) {
1433 		name_args = (struct fiodname_args *)data;
1434 		name = dev->si_name;
1435 		namlen = strlen(name) + 1;
1436 
1437 		devfs_debug(DEVFS_DEBUG_DEBUG,
1438 			    "ioctl, got: FIODNAME for %s\n", name);
1439 
1440 		if (namlen <= name_args->len)
1441 			error = copyout(dev->si_name, name_args->name, namlen);
1442 		else
1443 			error = EINVAL;
1444 
1445 		devfs_debug(DEVFS_DEBUG_DEBUG,
1446 			    "ioctl stuff: error: %d\n", error);
1447 		goto out;
1448 	}
1449 
1450 	error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
1451 
1452 #if 0
1453 	if (node) {
1454 		nanotime(&node->atime);
1455 		nanotime(&node->mtime);
1456 	}
1457 #endif
1458 	if (com == TIOCSCTTY) {
1459 		devfs_debug(DEVFS_DEBUG_DEBUG,
1460 			    "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1461 			    dev->si_name);
1462 	}
1463 	if (error == 0 && com == TIOCSCTTY) {
1464 		struct proc *p = curthread->td_proc;
1465 		struct session *sess;
1466 
1467 		devfs_debug(DEVFS_DEBUG_DEBUG,
1468 			    "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1469 			    dev->si_name);
1470 		if (p == NULL) {
1471 			error = ENOTTY;
1472 			goto out;
1473 		}
1474 		sess = p->p_session;
1475 
1476 		/*
1477 		 * Do nothing if reassigning same control tty
1478 		 */
1479 		if (sess->s_ttyvp == vp) {
1480 			error = 0;
1481 			goto out;
1482 		}
1483 
1484 		/*
1485 		 * Get rid of reference to old control tty
1486 		 */
1487 		ovp = sess->s_ttyvp;
1488 		vref(vp);
1489 		sess->s_ttyvp = vp;
1490 		if (ovp)
1491 			vrele(ovp);
1492 	}
1493 
1494 out:
1495 	release_dev(dev);
1496 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1497 	return (error);
1498 }
1499 
1500 
1501 static int
1502 devfs_spec_fsync(struct vop_fsync_args *ap)
1503 {
1504 	struct vnode *vp = ap->a_vp;
1505 	int error;
1506 
1507 	if (!vn_isdisk(vp, NULL))
1508 		return (0);
1509 
1510 	/*
1511 	 * Flush all dirty buffers associated with a block device.
1512 	 */
1513 	error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1514 	return (error);
1515 }
1516 
1517 static int
1518 devfs_spec_read(struct vop_read_args *ap)
1519 {
1520 	struct devfs_node *node;
1521 	struct vnode *vp;
1522 	struct uio *uio;
1523 	cdev_t dev;
1524 	int error;
1525 
1526 	vp = ap->a_vp;
1527 	dev = vp->v_rdev;
1528 	uio = ap->a_uio;
1529 	node = DEVFS_NODE(vp);
1530 
1531 	if (dev == NULL)		/* device was revoked */
1532 		return (EBADF);
1533 	if (uio->uio_resid == 0)
1534 		return (0);
1535 
1536 	vn_unlock(vp);
1537 	error = dev_dread(dev, uio, ap->a_ioflag);
1538 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1539 
1540 	if (node)
1541 		nanotime(&node->atime);
1542 
1543 	return (error);
1544 }
1545 
1546 /*
1547  * Vnode op for write
1548  *
1549  * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1550  *	      struct ucred *a_cred)
1551  */
1552 static int
1553 devfs_spec_write(struct vop_write_args *ap)
1554 {
1555 	struct devfs_node *node;
1556 	struct vnode *vp;
1557 	struct uio *uio;
1558 	cdev_t dev;
1559 	int error;
1560 
1561 	vp = ap->a_vp;
1562 	dev = vp->v_rdev;
1563 	uio = ap->a_uio;
1564 	node = DEVFS_NODE(vp);
1565 
1566 	KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1567 
1568 	if (dev == NULL)		/* device was revoked */
1569 		return (EBADF);
1570 
1571 	vn_unlock(vp);
1572 	error = dev_dwrite(dev, uio, ap->a_ioflag);
1573 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1574 
1575 	if (node) {
1576 		nanotime(&node->atime);
1577 		nanotime(&node->mtime);
1578 	}
1579 
1580 	return (error);
1581 }
1582 
1583 /*
1584  * Device ioctl operation.
1585  *
1586  * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1587  *	      int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1588  */
1589 static int
1590 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1591 {
1592 	struct vnode *vp = ap->a_vp;
1593 	struct devfs_node *node;
1594 	cdev_t dev;
1595 
1596 	if ((dev = vp->v_rdev) == NULL)
1597 		return (EBADF);		/* device was revoked */
1598 	node = DEVFS_NODE(vp);
1599 
1600 #if 0
1601 	if (node) {
1602 		nanotime(&node->atime);
1603 		nanotime(&node->mtime);
1604 	}
1605 #endif
1606 
1607 	return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1608 			   ap->a_cred, ap->a_sysmsg));
1609 }
1610 
1611 /*
1612  * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1613  */
1614 /* ARGSUSED */
1615 static int
1616 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1617 {
1618 	struct vnode *vp = ap->a_vp;
1619 	struct devfs_node *node;
1620 	cdev_t dev;
1621 
1622 	if ((dev = vp->v_rdev) == NULL)
1623 		return (EBADF);		/* device was revoked (EBADF) */
1624 	node = DEVFS_NODE(vp);
1625 
1626 #if 0
1627 	if (node)
1628 		nanotime(&node->atime);
1629 #endif
1630 
1631 	return (dev_dkqfilter(dev, ap->a_kn));
1632 }
1633 
1634 /*
1635  * Convert a vnode strategy call into a device strategy call.  Vnode strategy
1636  * calls are not limited to device DMA limits so we have to deal with the
1637  * case.
1638  *
1639  * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1640  */
1641 static int
1642 devfs_spec_strategy(struct vop_strategy_args *ap)
1643 {
1644 	struct bio *bio = ap->a_bio;
1645 	struct buf *bp = bio->bio_buf;
1646 	struct buf *nbp;
1647 	struct vnode *vp;
1648 	struct mount *mp;
1649 	int chunksize;
1650 	int maxiosize;
1651 
1652 	if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1653 		buf_start(bp);
1654 
1655 	/*
1656 	 * Collect statistics on synchronous and asynchronous read
1657 	 * and write counts for disks that have associated filesystems.
1658 	 */
1659 	vp = ap->a_vp;
1660 	KKASSERT(vp->v_rdev != NULL);	/* XXX */
1661 	if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1662 		if (bp->b_cmd == BUF_CMD_READ) {
1663 			if (bp->b_flags & BIO_SYNC)
1664 				mp->mnt_stat.f_syncreads++;
1665 			else
1666 				mp->mnt_stat.f_asyncreads++;
1667 		} else {
1668 			if (bp->b_flags & BIO_SYNC)
1669 				mp->mnt_stat.f_syncwrites++;
1670 			else
1671 				mp->mnt_stat.f_asyncwrites++;
1672 		}
1673 	}
1674 
1675         /*
1676          * Device iosize limitations only apply to read and write.  Shortcut
1677          * the I/O if it fits.
1678          */
1679 	if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1680 		devfs_debug(DEVFS_DEBUG_DEBUG,
1681 			    "%s: si_iosize_max not set!\n",
1682 			    dev_dname(vp->v_rdev));
1683 		maxiosize = MAXPHYS;
1684 	}
1685 #if SPEC_CHAIN_DEBUG & 2
1686 	maxiosize = 4096;
1687 #endif
1688         if (bp->b_bcount <= maxiosize ||
1689             (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1690                 dev_dstrategy_chain(vp->v_rdev, bio);
1691                 return (0);
1692         }
1693 
1694 	/*
1695 	 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1696 	 */
1697 	nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1698 	initbufbio(nbp);
1699 	buf_dep_init(nbp);
1700 	BUF_LOCK(nbp, LK_EXCLUSIVE);
1701 	BUF_KERNPROC(nbp);
1702 	nbp->b_vp = vp;
1703 	nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1704 	nbp->b_data = bp->b_data;
1705 	nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1706 	nbp->b_bio1.bio_offset = bio->bio_offset;
1707 	nbp->b_bio1.bio_caller_info1.ptr = bio;
1708 
1709 	/*
1710 	 * Start the first transfer
1711 	 */
1712 	if (vn_isdisk(vp, NULL))
1713 		chunksize = vp->v_rdev->si_bsize_phys;
1714 	else
1715 		chunksize = DEV_BSIZE;
1716 	chunksize = maxiosize / chunksize * chunksize;
1717 #if SPEC_CHAIN_DEBUG & 1
1718 	devfs_debug(DEVFS_DEBUG_DEBUG,
1719 		    "spec_strategy chained I/O chunksize=%d\n",
1720 		    chunksize);
1721 #endif
1722 	nbp->b_cmd = bp->b_cmd;
1723 	nbp->b_bcount = chunksize;
1724 	nbp->b_bufsize = chunksize;	/* used to detect a short I/O */
1725 	nbp->b_bio1.bio_caller_info2.index = chunksize;
1726 
1727 #if SPEC_CHAIN_DEBUG & 1
1728 	devfs_debug(DEVFS_DEBUG_DEBUG,
1729 		    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1730 		    bp, 0, bp->b_bcount, nbp->b_bcount);
1731 #endif
1732 
1733 	dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1734 
1735 	if (DEVFS_NODE(vp)) {
1736 		nanotime(&DEVFS_NODE(vp)->atime);
1737 		nanotime(&DEVFS_NODE(vp)->mtime);
1738 	}
1739 
1740 	return (0);
1741 }
1742 
1743 /*
1744  * Chunked up transfer completion routine - chain transfers until done
1745  *
1746  * NOTE: MPSAFE callback.
1747  */
1748 static
1749 void
1750 devfs_spec_strategy_done(struct bio *nbio)
1751 {
1752 	struct buf *nbp = nbio->bio_buf;
1753 	struct bio *bio = nbio->bio_caller_info1.ptr;	/* original bio */
1754 	struct buf *bp = bio->bio_buf;			/* original bp */
1755 	int chunksize = nbio->bio_caller_info2.index;	/* chunking */
1756 	int boffset = nbp->b_data - bp->b_data;
1757 
1758 	if (nbp->b_flags & B_ERROR) {
1759 		/*
1760 		 * An error terminates the chain, propogate the error back
1761 		 * to the original bp
1762 		 */
1763 		bp->b_flags |= B_ERROR;
1764 		bp->b_error = nbp->b_error;
1765 		bp->b_resid = bp->b_bcount - boffset +
1766 			      (nbp->b_bcount - nbp->b_resid);
1767 #if SPEC_CHAIN_DEBUG & 1
1768 		devfs_debug(DEVFS_DEBUG_DEBUG,
1769 			    "spec_strategy: chain %p error %d bcount %d/%d\n",
1770 			    bp, bp->b_error, bp->b_bcount,
1771 			    bp->b_bcount - bp->b_resid);
1772 #endif
1773 	} else if (nbp->b_resid) {
1774 		/*
1775 		 * A short read or write terminates the chain
1776 		 */
1777 		bp->b_error = nbp->b_error;
1778 		bp->b_resid = bp->b_bcount - boffset +
1779 			      (nbp->b_bcount - nbp->b_resid);
1780 #if SPEC_CHAIN_DEBUG & 1
1781 		devfs_debug(DEVFS_DEBUG_DEBUG,
1782 			    "spec_strategy: chain %p short read(1) "
1783 			    "bcount %d/%d\n",
1784 			    bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1785 #endif
1786 	} else if (nbp->b_bcount != nbp->b_bufsize) {
1787 		/*
1788 		 * A short read or write can also occur by truncating b_bcount
1789 		 */
1790 #if SPEC_CHAIN_DEBUG & 1
1791 		devfs_debug(DEVFS_DEBUG_DEBUG,
1792 			    "spec_strategy: chain %p short read(2) "
1793 			    "bcount %d/%d\n",
1794 			    bp, nbp->b_bcount + boffset, bp->b_bcount);
1795 #endif
1796 		bp->b_error = 0;
1797 		bp->b_bcount = nbp->b_bcount + boffset;
1798 		bp->b_resid = nbp->b_resid;
1799 	} else if (nbp->b_bcount + boffset == bp->b_bcount) {
1800 		/*
1801 		 * No more data terminates the chain
1802 		 */
1803 #if SPEC_CHAIN_DEBUG & 1
1804 		devfs_debug(DEVFS_DEBUG_DEBUG,
1805 			    "spec_strategy: chain %p finished bcount %d\n",
1806 			    bp, bp->b_bcount);
1807 #endif
1808 		bp->b_error = 0;
1809 		bp->b_resid = 0;
1810 	} else {
1811 		/*
1812 		 * Continue the chain
1813 		 */
1814 		boffset += nbp->b_bcount;
1815 		nbp->b_data = bp->b_data + boffset;
1816 		nbp->b_bcount = bp->b_bcount - boffset;
1817 		if (nbp->b_bcount > chunksize)
1818 			nbp->b_bcount = chunksize;
1819 		nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1820 		nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1821 
1822 #if SPEC_CHAIN_DEBUG & 1
1823 		devfs_debug(DEVFS_DEBUG_DEBUG,
1824 			    "spec_strategy: chain %p offset %d/%d bcount %d\n",
1825 			    bp, boffset, bp->b_bcount, nbp->b_bcount);
1826 #endif
1827 
1828 		dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1829 		return;
1830 	}
1831 
1832 	/*
1833 	 * Fall through to here on termination.  biodone(bp) and
1834 	 * clean up and free nbp.
1835 	 */
1836 	biodone(bio);
1837 	BUF_UNLOCK(nbp);
1838 	uninitbufbio(nbp);
1839 	kfree(nbp, M_DEVBUF);
1840 }
1841 
1842 /*
1843  * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1844  */
1845 static int
1846 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1847 {
1848 	struct buf *bp;
1849 
1850 	/*
1851 	 * XXX: This assumes that strategy does the deed right away.
1852 	 * XXX: this may not be TRTTD.
1853 	 */
1854 	KKASSERT(ap->a_vp->v_rdev != NULL);
1855 	if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
1856 		return (0);
1857 	bp = geteblk(ap->a_length);
1858 	bp->b_cmd = BUF_CMD_FREEBLKS;
1859 	bp->b_bio1.bio_offset = ap->a_offset;
1860 	bp->b_bcount = ap->a_length;
1861 	dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1862 	return (0);
1863 }
1864 
1865 /*
1866  * Implement degenerate case where the block requested is the block
1867  * returned, and assume that the entire device is contiguous in regards
1868  * to the contiguous block range (runp and runb).
1869  *
1870  * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1871  *	     off_t *a_doffsetp, int *a_runp, int *a_runb)
1872  */
1873 static int
1874 devfs_spec_bmap(struct vop_bmap_args *ap)
1875 {
1876 	if (ap->a_doffsetp != NULL)
1877 		*ap->a_doffsetp = ap->a_loffset;
1878 	if (ap->a_runp != NULL)
1879 		*ap->a_runp = MAXBSIZE;
1880 	if (ap->a_runb != NULL) {
1881 		if (ap->a_loffset < MAXBSIZE)
1882 			*ap->a_runb = (int)ap->a_loffset;
1883 		else
1884 			*ap->a_runb = MAXBSIZE;
1885 	}
1886 	return (0);
1887 }
1888 
1889 
1890 /*
1891  * Special device advisory byte-level locks.
1892  *
1893  * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1894  *		struct flock *a_fl, int a_flags)
1895  */
1896 /* ARGSUSED */
1897 static int
1898 devfs_spec_advlock(struct vop_advlock_args *ap)
1899 {
1900 	return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1901 }
1902 
1903 /*
1904  * NOTE: MPSAFE callback.
1905  */
1906 static void
1907 devfs_spec_getpages_iodone(struct bio *bio)
1908 {
1909 	bio->bio_buf->b_cmd = BUF_CMD_DONE;
1910 	wakeup(bio->bio_buf);
1911 }
1912 
1913 /*
1914  * spec_getpages() - get pages associated with device vnode.
1915  *
1916  * Note that spec_read and spec_write do not use the buffer cache, so we
1917  * must fully implement getpages here.
1918  */
1919 static int
1920 devfs_spec_getpages(struct vop_getpages_args *ap)
1921 {
1922 	vm_offset_t kva;
1923 	int error;
1924 	int i, pcount, size;
1925 	struct buf *bp;
1926 	vm_page_t m;
1927 	vm_ooffset_t offset;
1928 	int toff, nextoff, nread;
1929 	struct vnode *vp = ap->a_vp;
1930 	int blksiz;
1931 	int gotreqpage;
1932 
1933 	error = 0;
1934 	pcount = round_page(ap->a_count) / PAGE_SIZE;
1935 
1936 	/*
1937 	 * Calculate the offset of the transfer and do sanity check.
1938 	 */
1939 	offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1940 
1941 	/*
1942 	 * Round up physical size for real devices.  We cannot round using
1943 	 * v_mount's block size data because v_mount has nothing to do with
1944 	 * the device.  i.e. it's usually '/dev'.  We need the physical block
1945 	 * size for the device itself.
1946 	 *
1947 	 * We can't use v_rdev->si_mountpoint because it only exists when the
1948 	 * block device is mounted.  However, we can use v_rdev.
1949 	 */
1950 	if (vn_isdisk(vp, NULL))
1951 		blksiz = vp->v_rdev->si_bsize_phys;
1952 	else
1953 		blksiz = DEV_BSIZE;
1954 
1955 	size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
1956 
1957 	bp = getpbuf_kva(NULL);
1958 	kva = (vm_offset_t)bp->b_data;
1959 
1960 	/*
1961 	 * Map the pages to be read into the kva.
1962 	 */
1963 	pmap_qenter(kva, ap->a_m, pcount);
1964 
1965 	/* Build a minimal buffer header. */
1966 	bp->b_cmd = BUF_CMD_READ;
1967 	bp->b_bcount = size;
1968 	bp->b_resid = 0;
1969 	bsetrunningbufspace(bp, size);
1970 
1971 	bp->b_bio1.bio_offset = offset;
1972 	bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
1973 
1974 	mycpu->gd_cnt.v_vnodein++;
1975 	mycpu->gd_cnt.v_vnodepgsin += pcount;
1976 
1977 	/* Do the input. */
1978 	vn_strategy(ap->a_vp, &bp->b_bio1);
1979 
1980 	crit_enter();
1981 
1982 	/* We definitely need to be at splbio here. */
1983 	while (bp->b_cmd != BUF_CMD_DONE)
1984 		tsleep(bp, 0, "spread", 0);
1985 
1986 	crit_exit();
1987 
1988 	if (bp->b_flags & B_ERROR) {
1989 		if (bp->b_error)
1990 			error = bp->b_error;
1991 		else
1992 			error = EIO;
1993 	}
1994 
1995 	/*
1996 	 * If EOF is encountered we must zero-extend the result in order
1997 	 * to ensure that the page does not contain garabge.  When no
1998 	 * error occurs, an early EOF is indicated if b_bcount got truncated.
1999 	 * b_resid is relative to b_bcount and should be 0, but some devices
2000 	 * might indicate an EOF with b_resid instead of truncating b_bcount.
2001 	 */
2002 	nread = bp->b_bcount - bp->b_resid;
2003 	if (nread < ap->a_count)
2004 		bzero((caddr_t)kva + nread, ap->a_count - nread);
2005 	pmap_qremove(kva, pcount);
2006 
2007 	gotreqpage = 0;
2008 	for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2009 		nextoff = toff + PAGE_SIZE;
2010 		m = ap->a_m[i];
2011 
2012 		m->flags &= ~PG_ZERO;
2013 
2014 		/*
2015 		 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2016 		 *	 pmap modified bit.  pmap modified bit should have
2017 		 *	 already been cleared.
2018 		 */
2019 		if (nextoff <= nread) {
2020 			m->valid = VM_PAGE_BITS_ALL;
2021 			vm_page_undirty(m);
2022 		} else if (toff < nread) {
2023 			/*
2024 			 * Since this is a VM request, we have to supply the
2025 			 * unaligned offset to allow vm_page_set_valid()
2026 			 * to zero sub-DEV_BSIZE'd portions of the page.
2027 			 */
2028 			vm_page_set_valid(m, 0, nread - toff);
2029 			vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2030 		} else {
2031 			m->valid = 0;
2032 			vm_page_undirty(m);
2033 		}
2034 
2035 		if (i != ap->a_reqpage) {
2036 			/*
2037 			 * Just in case someone was asking for this page we
2038 			 * now tell them that it is ok to use.
2039 			 */
2040 			if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2041 				if (m->valid) {
2042 					if (m->flags & PG_REFERENCED) {
2043 						vm_page_activate(m);
2044 					} else {
2045 						vm_page_deactivate(m);
2046 					}
2047 					vm_page_wakeup(m);
2048 				} else {
2049 					vm_page_free(m);
2050 				}
2051 			} else {
2052 				vm_page_free(m);
2053 			}
2054 		} else if (m->valid) {
2055 			gotreqpage = 1;
2056 			/*
2057 			 * Since this is a VM request, we need to make the
2058 			 * entire page presentable by zeroing invalid sections.
2059 			 */
2060 			if (m->valid != VM_PAGE_BITS_ALL)
2061 			    vm_page_zero_invalid(m, FALSE);
2062 		}
2063 	}
2064 	if (!gotreqpage) {
2065 		m = ap->a_m[ap->a_reqpage];
2066 		devfs_debug(DEVFS_DEBUG_WARNING,
2067 	    "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2068 			devtoname(vp->v_rdev), error, bp, bp->b_vp);
2069 		devfs_debug(DEVFS_DEBUG_WARNING,
2070 	    "               size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2071 		    size, bp->b_resid, ap->a_count, m->valid);
2072 		devfs_debug(DEVFS_DEBUG_WARNING,
2073 	    "               nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2074 		    nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2075 		/*
2076 		 * Free the buffer header back to the swap buffer pool.
2077 		 */
2078 		relpbuf(bp, NULL);
2079 		return VM_PAGER_ERROR;
2080 	}
2081 	/*
2082 	 * Free the buffer header back to the swap buffer pool.
2083 	 */
2084 	relpbuf(bp, NULL);
2085 	if (DEVFS_NODE(ap->a_vp))
2086 		nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2087 	return VM_PAGER_OK;
2088 }
2089 
2090 static __inline
2091 int
2092 sequential_heuristic(struct uio *uio, struct file *fp)
2093 {
2094 	/*
2095 	 * Sequential heuristic - detect sequential operation
2096 	 */
2097 	if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2098 	    uio->uio_offset == fp->f_nextoff) {
2099 		/*
2100 		 * XXX we assume that the filesystem block size is
2101 		 * the default.  Not true, but still gives us a pretty
2102 		 * good indicator of how sequential the read operations
2103 		 * are.
2104 		 */
2105 		int tmpseq = fp->f_seqcount;
2106 
2107 		tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2108 		if (tmpseq > IO_SEQMAX)
2109 			tmpseq = IO_SEQMAX;
2110 		fp->f_seqcount = tmpseq;
2111 		return(fp->f_seqcount << IO_SEQSHIFT);
2112 	}
2113 
2114 	/*
2115 	 * Not sequential, quick draw-down of seqcount
2116 	 */
2117 	if (fp->f_seqcount > 1)
2118 		fp->f_seqcount = 1;
2119 	else
2120 		fp->f_seqcount = 0;
2121 	return(0);
2122 }
2123